Data Mining Techniques for Segmentation - Data Mining Techniques in CRM: Inside Customer Segmentation - page 120

with balanced proportions of multiple clusters. These records should be
examined separately. The respective scoring rules would have low confidence
values, indicating a large level of uncertainty in cluster assignment. Users
aiming at a more solid and consistent clustering solution might decide not
to use these rules for cluster scoring and classify the relevant records as an
''unidentified'' cluster.

Figure 3.27 shows the first two levels of a decision tree model that was built to
profile the cluster membership field of our telecommunications example presented
in ''Understanding the Clusters through Profiling''.
The number of voice calls is selected for the first split. Customers with high
voice usage are further partitioned according to the total number of roaming calls.

cluster-1

$T-Twostep

cluster-2
cluster-3
cluster-4
cluster-5
cluster-6

Node 0

−

VOICE_OUT_CALLS
Improvement = 0.142

< = 219.083 > 219.083

Node 1 Node 2

− −

SMS_OUT_CALLS OUT_CALLS_ROAMING

Improvement = 0.148 Improvement = 0.013

< = 36.083 > 36.083 < = 22.417 > 22.417

Node 3 Node 4 Node 5 Node 6

+ + +

Figure 3.27 A decision tree model applied for profiling the clusters.